Last week we were impressed with some of the AI announcements and demonstrations at Google’s I/O developer conference, most notably the very lifelike phone calls made by Google’s assistant, placing both a hair appointment and a dinner appointment. It was so lifelike it sparked a stinging rebuttal by several people concerned with ethics. Now, Google says it’ll have the bot identify itself when it places these types of phone calls. If you haven’t already seen it, watch the calls here:
Now let’s take a look at some other AI and ML applications that were overshadowed by Google’s high-profile event.
Rockin’ the treetops
Researchers in Japan have created an innovative, less expensive way to classify trees. They send up a drone to take aerial shots of land sections in question and then use deep learning to classify the types of trees based on their tree crowns. With 89 percent accuracy, this algorithm divided the tree tops into seven categories. If this turns out to be a viable alternative to typical large-scale tree classification methods, they say it could save scientists and land managers quite a bit of money.
My specialized grid cells that fire in hexagonal patterns will guide me there
Mammals have something in their brains called grid cells that helps them navigate. We’re still learning about these cells because they’re a relatively new discovery (they were discovered in 2005), but we do know that they create hexagon-shaped patterns, are located in the brain, and help mammals navigate. But we don’t know how they aid navigation. Interestingly enough, as Google’s DeepMind tried to answer an aspect of this question, researchers unexpectedly discovered that the deep learning algorithms developed a very similar system to the human grid cells while trying to solve navigation problems.
No way, Dengue.
Dengue kills a lot of people annually, and one man wants to use machine learning to stop it. He’s employing a system that may be able to predict dengue outbreaks as far out as three months ahead of time, and it’s being rolled out in Malaysia. It works like this: doctors report dengue cases, this data is fed into the system, the system then searches over 90 databases for hundreds of variables that would affect the disease’s spread, and then from there, the system predicts where the next outbreak will be within 400 meters. It has an 81 to 84 percent accuracy, so there’s hope that it will help officials get ahead of the disease and better direct resources.
This isn’t the first time we’ve heard of something like this. When we interviewed a Tableau exec for our podcast, we learned about a similar initiative in Zambia with Malaria that has seen a lot of success.
Chatterbaby attempts to translate your baby’s cries
Making 3D motion capture easier
Astronomers get closer to more easily identifying gravitational lenses
The way we move our eyes may predict our personalities
Scientists use ML to identify and track whale calls with 98 percent accuracy
Deep learning technique helps correct poorly lit photos